Thompson Sampling: Predicting Behavior in Games and Markets

Abstract

This paper proposes Thompson Sampling as a unifying and tractable theory of expectation formation, which is in line with theories of the brain. Thompson Sampling means that in uncertain environments, agents update their beliefs in a Bayesian way, and subsequently make a random draw from the posterior. Conditional on that random draw, agents optimize. Thompson Sampling helps explain data from experimental games, market experiments, and survey data on inflation expectations in a unified fashion. In comparison to other modeling approaches, Thompson Sampling stands out in terms of statistical fit and predictive accuracy.

Keywords: Learning, bounded rationality, behavioral game theory, expectations, stochastic choice

JEL Classification: C91, C92, D84, E37